Predict the residue type of a spin system and the optimal arrangement of residue types

It is possible in Analysis to predict the residue type of a spin system through the spin system typing functionality, accessible through M: Assignment: Spin System Typing.

The Assignment: Spin System Typing popup uses chemical shift information, obtained from the resonances of a spin system group, to predict which kind of residue a spin system could be. Naturally, the more resonances/shifts there are in a spin system the better the prediction will be. Predictions are either made for a single spin system in isolation, or for a whole chain; by shuffling the residue to spin system mapping to give the optimum arrangement.

Predicting an individual Spin System

Select a spin system in the ‘Spin Systems’ table and click [Show Individual Classification] to predict the residue type for the spin system. This prediction is covered in the Spin System Type Scores popup and is further explained here.

Predicting the optimal arrangement of Residue Types

This functionality attempts to find the best match of spin system to residue type by performing a Monte Carlo search to arrange each group of chemical shifts amongst the residue types found in the chain. The main principle is that the chain’s residues dictate how many spin systems of a given type may be found, such that for example if there is only one Threonine residue then only one spin system may be predicted to be of Threonine type. This result comes naturally from shuffling the spin systems, with their chemical shifts, amongst the residue slots (disregarding sequence position).

To predict residue types in the whole chain, do as follows:

• • Set “Chain:” and “Shift List:” to use.

  • Set “Isotope shifts to consider:” and choose various Monte Carlo search options. Increasing the number of search steps in “Num Search Steps” may help if the search does not converge. Increasing “Ensemble size:” (how many test mappings are optimised at the same time) may help if the prediction gets stuck in local minima; different runs predict different arrangement, but larger ensembles require more search steps to converge.

  • Click [Run Typing].

The prediction may take some time to run, depending upon the number of residues and spin systems that are being matched, but gives a graphical output of the progress:

If the final prediction looks good the Highest Scoring Mappings popup may be closed and [Assign Types] may be used to set the residue types for all of the spin systems in the main table that match only a single type and have a score above the assignment threshold value. Spin systems that already have a type or full residue assignment will not be affected.

Overall, it should be noted that if a human being cannot readily predict the probably types of a spin system from its shifts alone, then this search tool cannot be expected to do a good job; it is merely an optimiser to address the problem of shuffling spin systems within a chain.

The score values under |Score| are unnormalised log-odds values and are not especially meaningful in the human sense, other than higher is better (closer to zero for negative values). This issue will be addressed in the future. Spin systems without a unique type prediction will not get a final score.